Interative approaches to convex feasibility problems.

Abstract

Solutions to convex feasibility problems are generally found by iteratively constructing sequences that converge strongly or weakly to it. In this study, four types of iteration schemes are considered in an attempt to find a point in the intersection of some closed and convex sets. The iteration scheme Xn+l = (1 - λn+1)y + λn+1Tn+lxn is first considered for infinitely many nonexpansive maps Tl , T2 , T3 , ... in a Hilbert space. A result of Shimizu and Takahashi [33] is generalized, and it is shown that the sequence of iterates converge to Py, where P is some projection. This is further generalized to a uniformly smooth Banach space having a weakly continuous duality map. Here the iterates converge to Qy, where Q is a sunny nonexpansive retraction. For this same iteration scheme, with finitely many maps Tl , T2, ... , TN , a complementary result to a result of Bauschke [2] is proved by introducing a new condition on the sequence of parameters (λn). The iterates converge to Py, where P is the projection onto the intersection of the fixed point sets of the Tis. Both this result and Bauschke's result [2] are then generalized to a uniformly smooth Banach space, and to a reflexive Banach space having a weakly continuous duality map and having Reich's property. Now the iterates converge to Qy, where Q is the unique sunny nonexpansive retraction onto the intersection of the fixed point sets of the Tis. For a random map r : N  {I, 2, ... ,N}, the iteration scheme xn+l = Tr(n+l)xn is considered. In a finite dimensional Hilbert space with Tr(n) = Pr(n) , the iterates converge to a point in the intersection of the fixed point sets of the PiS. In an arbitrary Banach space, under certain conditions on the mappings, the iterates converge to a point in the intersection of the fixed point sets of the Tis. For the scheme xn+l = (1- λn+l)xn+λn+lTr(n+l)xn, in a finite dimensional Hilbert space the iterates converge to a point in the intersection of the fixed point sets of the Tis, and in an infinite dimensional Hilbert space with the added assumption that the random map r is quasi-cyclic, then the iterates converge weakly to a point in the intersection of the fixed point sets of the Tis. Lastly, the minimization of a convex function θ is considered over some closed and convex subset of a Hilbert space. For both the case where θ is a quadratic function and for the general case, first the unique fixed points of some maps Tλ are shown to converge to the unique minimizer of θ and then an algorithm is proposed that converges to this unique minimizer.